Scientific writing 3.0: A reader and writer's guide - Jean-Luc Lebrun, Justin Lebrun 2021

Reasons for betrayed expectations
Detect sentence fluidity problems
The reading toolkit

Unannounced topic change and ambiguous pronoun

The images in our set are normalized to have a consistent gray level when output. The coordinates of the eyes are automatically detected and set to a fixed position. They are then resampled to a given size. After normalization…

The topic of the first sentence suggests two possible candidates for the next sentence:

1. The benefit of having a consistent gray level

2. Additional normalization step on the images, such as image contrast

The “eye” topic of sentence 2 does not fit expectation 1 or 2. It could have, had the writer specified that the images were face images. Only after readers finish reading sentence 2, do they realize that it ties in with candidate 2 — the additional normalization step. Based on this, readers expect two new candidates for the next sentence:

1. Yet another normalization step

2. A reason for fixing the eye coordinates, or the location of the fixed position

The pronoun ’they’ starts the next sentence. Readers think the pronoun refers to ’the coordinates of the eyes’, and they expect a reason for fixing these coordinates. But they reach an impasse when the verb “resampled” arrives because coordinates are unlikely to be resampled. Readers briefly stop and re-read to discover that ’they’ refers to the images mentioned in sentence 1. The pronoun was ambiguous. The third sentence was simply another normalization step (expectation 1).

Here is a possible rewrite.

All face images are normalized (1) to display consistent gray levels, (2) to automatically place the eyes at the same location in each image, and finally (3) to set each image to a 128 × 128 size through resampling. After normalization, …

Text precision is increased. The sentence is long, but precise and easy to follow. The enumeration and the parallel syntax set the expectations that we are moving from one normalization step to the next.

Adjectival or adverbial claim not followed by evidence

Adjectives or adverbs are words that set expectations in three ways:

· They make a claim that requires validation, as in “The world is flat”.

· They state an existing (often well-known) situation judged unsatisfactory that requires correction, as is “Airplane engine maintenance is very labor-intensive”. In this case, very labor-intensive sets the expectation for its opposite, less labor-intensive. Note the role of very as a judgmental word to enhance the expectation for a more satisfactory solution.

· They state a situation in a negative or a pejorative light that requires the need for a change towards the positive or the ameliorative as in “trapping is unimportant at high temperatures.” The sentence sets the expectations that trapping may be important at low temperatures.

Drop test simulations that rely on a fine mesh of 3D finite elements are very time-consuming. They are required to save the time and cost involved in actual physical tests conducted at board level. We propose a simplified model that considers the board as a beam structure…

The adjective time-consuming makes a claim associated to the Finite Element Method. Readers in that field are already aware that fine mesh 3D modeling is very time-consuming. They expect the writer to present faster methods in the next sentence. It comes… in the third sentence. You may object that the writer uses a good progression scheme, the constant topic progression: ’drop test simulations’, and the pronoun ’They,’ although the pronoun is ambiguous because it could also replace finite elements. To answer the objection, it is worth emphasizing that expectations import more than progression. To meet the expectations of the reader, these sentences must be rewritten. Here is a possible rewrite.

Simulations based on the analysis of a fine 3D mesh of finite elements save the cost, but not the time involved in conducting physical board level drop tests. Saving time is possible if the model is simplified. We propose a model that simplifies the board by making it a beam structure…

In this rewrite, chain progression is set, and expectations are met.

Unclear answers to clear questions

Beyond which threshold value should the degree of asymmetry of the brain be considered abnormal? The answer depends on what kind of brain information is required. When one studies pathological abnormality, false positives and false negatives should be kept low. The threshold value differentiating normal from abnormal asymmetry could be estimated from patient data, but how sensitive or how specific that value would be, is unclear.

The starting question sets an expectation for a precise numerical value. The it-depends answer is unclear and disappoints. Indeed, the second sentence acts as a second indirect question: What is the brain information required? Again, the answer disappoints, because the writer gives an unexpected answer that reveals a concern for the statistical quality of the data. The reader could not possibly have guessed that, and is therefore surprised by the false positives/false negatives sentence. At this point, the reader usually gives up and, having been surprised twice in a row, takes a neutral attitude. The writer then tells the reader that a threshold value is of little use if it is not sensitive or specific. Here is a possible rewrite:

Assuming that the degree of brain asymmetry is a reliable indicator of brain pathology, can a diagnostic of abnormality be based on an asymmetry threshold value? To answer that question, it will be necessary to determine whether such a value derived from patient data has enough sensitivity and specificity.

Unjustified choice

“The methods for cursor control [Brain Computer Interface] come under two categories, regression [references] and classification [more references]. Each of them has its merits. In our study we adopt the classification method,...”⁴ (the text that follows describes the method, but does not give the reason for the choice).

The writer presents the choice he faced: regression or classification. Both have their own advantages. The reader has two possible expectations:

· The writer will give the advantages and disadvantages of each. Obviously, the writer did not intend to do that. It would be better to remove the sentence “Each of them has its merits” because it creates the wrong expectations.

· The writer will choose one category of methods and justify it.

Which disadvantage, or which advantage led the writer to prefer the classification method? The writer does not say. Note that even if the second sentence “Each of them has its merits” is removed, the reader will still want to know why the writer chose the classification methods over the regression methods. Whenever you announce and make a choice, the reader wants to know why, even if the choice is arbitrary.

The methods for cursor control [Brain Computer Interface] come under two categories, regression [references] and classification [more references]. We decided to adopt the classification method because…

Broken repeated patterns and paraphrase

The emergence of Hidden Markov Models (HMM) in the 1970s allowed tremendous progress in speech recognition. HMMs are still the de-facto method for speech recognition today. However, some argue that HMMs are not a panacea. At the same time, today’s speech recognition systems are far smarter than those in the earlier days.

The first sentence establishes the paragraph topic: HMMs. It places it in a good light. The reader expects the praise to continue, and it does in sentence 2. The reader is given two expectations for sentence 3: (a) examples of the current use of HMMs in speech recognition, (b) or the reasons for the continued use of HMMs (maybe recent improvements to the basic HMM method).

In sentence 3, the writer instead surprises the reader by announcing the limitations. Expectations reset, the reader now expects the writer to name a problem where HMMs do not excel. Unfortunately, in sentence 4, the writer seems to change topic. “At the same time” is a phrase similar to “in addition”: it masks the lack of a proper transition.

When breaking a pattern, it is better to announce the break in the same sentence as shown below.

Although HMMs are still the de-facto method for speech recognition today, they are not a panacea. As speech recognition applications increase in sophistication (automatic language recognition, speaker authentication), HMMs need to blend harmoniously with other statistical methods, in realtime.

Lack of knowledge (or unexplained word) and synonyms

Blood vessels carry blood cells as well as platelets, which are not as numerous as blood cells (ratio of 1 to 20). Platelets, besides being a source of growth factors, are also directly involved in homeostasis by aggregating to form a platelet plug that stops the bleeding. A thrombocyte count is usually included in a blood test as it is a means to determine diseases such as leukemia.

The third sentence blocks the non-expert reader who does not know that thrombocyte is synonymous with platelet. In this case, the writer did not wish to repeat platelet because it had already been mentioned twice in the previous sentence. Filling a paper with synonyms increases the knowledge gap and increases the memory load. As a rule, settle on one keyword (the simpler one) when two synonyms are available, and use it consistently in your paper.

A platelet count is usually included in a blood test as it is a means to determine diseases such as leukemia.

But if you wish to introduce a new keyword, define it in a just-in-time fashion as shown below.

A platelet count (also known as a thrombocyte count) is usually included in a blood test as it is a means to determine diseases such as leukemia.

¹ Qibin Sun and Shih-Fu Chang, A Robust and Secure Media Signature Scheme for JPEG Images, Journal of VLSI Signal Processing, Special Issue on MMSP2002, pp. 306—317, Vol. 41, No. 3, Nov., 2005.

² Papadias Dimitris, Tao Yufei, Lian Xiang, Xiao Xiaokui The VLDB Journal: the international journal on very large data bases, July 2007, v.16, no. 3, pp. 293—316.

³ Boon Ping Chan, Junhu Wei, Xiaoguang Wang, (2003) Synchron, Synchronization and management of shared state in hla-based distributed simulation, Proceedings of the Winter Simulation Conference,S. Chick, P. J. Sánchez, D. Ferrin, and D. J. Morrice, eds.

⁴ Zhu X, Guan C, Wu J, Cheng Y, Wang Y. (2005) Bayesian Method for Continuous Cursor Control in EEG-Based Brain-Computer Interface. Conf Proc IEEE Eng Med Biol Soc 7: 7052—7055.